[ovs-dev] [PATCH v16 5/8] dpif-offload-netlink: Implement dpif-offload-provider API

Chris Mi cmi at nvidia.com
Tue Oct 12 08:19:34 UTC 2021


Implement dpif-offload API for netlink datapath.

Signed-off-by: Chris Mi <cmi at nvidia.com>
Reviewed-by: Eli Britstein <elibr at nvidia.com>
---
 lib/automake.mk             |   1 +
 lib/dpif-netlink.c          |   2 +-
 lib/dpif-offload-netlink.c  | 209 ++++++++++++++++++++++++++++++++++++
 lib/dpif-offload-provider.h |  13 ++-
 4 files changed, 223 insertions(+), 2 deletions(-)
 create mode 100644 lib/dpif-offload-netlink.c

diff --git a/lib/automake.mk b/lib/automake.mk
index 9259f57de..18cffa827 100644
--- a/lib/automake.mk
+++ b/lib/automake.mk
@@ -446,6 +446,7 @@ lib_libopenvswitch_la_SOURCES += \
 	lib/dpif-netlink.h \
 	lib/dpif-netlink-rtnl.c \
 	lib/dpif-netlink-rtnl.h \
+	lib/dpif-offload-netlink.c \
 	lib/if-notifier.c \
 	lib/netdev-linux.c \
 	lib/netdev-linux.h \
diff --git a/lib/dpif-netlink.c b/lib/dpif-netlink.c
index f546b48e5..19ae543e6 100644
--- a/lib/dpif-netlink.c
+++ b/lib/dpif-netlink.c
@@ -4341,7 +4341,7 @@ const struct dpif_class dpif_netlink_class = {
     NULL,                       /* bond_add */
     NULL,                       /* bond_del */
     NULL,                       /* bond_stats_get */
-    NULL,                       /* dpif_offload_api */
+    &dpif_offload_netlink,
 };
 
 static int
diff --git a/lib/dpif-offload-netlink.c b/lib/dpif-offload-netlink.c
new file mode 100644
index 000000000..6ebf1aa0c
--- /dev/null
+++ b/lib/dpif-offload-netlink.c
@@ -0,0 +1,209 @@
+/*
+ * Copyright (c) 2021, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <config.h>
+#include <errno.h>
+#include <linux/psample.h>
+#include <sys/poll.h>
+
+#include "dpif-offload-provider.h"
+#include "netdev-offload.h"
+#include "netlink-protocol.h"
+#include "netlink-socket.h"
+#include "openvswitch/vlog.h"
+
+VLOG_DEFINE_THIS_MODULE(dpif_offload_netlink);
+
+static struct nl_sock *psample_sock;
+static int psample_family;
+
+/* Receive psample netlink message and save the attributes. */
+struct offload_psample {
+    struct nlattr *packet;      /* Packet data. */
+    int dp_group_id;            /* Mapping id for sFlow offload. */
+    int iifindex;               /* Input ifindex. */
+};
+
+/* In order to keep compatibility with kernels without psample module,
+ * return success even if psample is not initialized successfully. */
+static int
+dpif_offload_netlink_init(void)
+{
+    unsigned int psample_mcgroup;
+    int err;
+
+    if (!netdev_is_flow_api_enabled()) {
+        VLOG_DBG("Flow API is not enabled.");
+        return 0;
+    }
+
+    if (psample_sock) {
+        VLOG_DBG("Psample socket is already initialized.");
+        return 0;
+    }
+
+    err = nl_lookup_genl_family(PSAMPLE_GENL_NAME,
+                                &psample_family);
+    if (err) {
+        VLOG_WARN("Generic Netlink family '%s' does not exist: %s\n"
+                  "Please make sure the kernel module psample is loaded.",
+                  PSAMPLE_GENL_NAME, ovs_strerror(err));
+        return 0;
+    }
+
+    err = nl_lookup_genl_mcgroup(PSAMPLE_GENL_NAME,
+                                 PSAMPLE_NL_MCGRP_SAMPLE_NAME,
+                                 &psample_mcgroup);
+    if (err) {
+        VLOG_WARN("Failed to join Netlink multicast group '%s': %s",
+                  PSAMPLE_NL_MCGRP_SAMPLE_NAME, ovs_strerror(err));
+        return 0;
+    }
+
+    err = nl_sock_create(NETLINK_GENERIC, &psample_sock);
+    if (err) {
+        VLOG_WARN("Failed to create psample socket: %s", ovs_strerror(err));
+        return 0;
+    }
+
+    err = nl_sock_join_mcgroup(psample_sock, psample_mcgroup);
+    if (err) {
+        VLOG_WARN("Failed to join psample mcgroup: %s", ovs_strerror(err));
+        nl_sock_destroy(psample_sock);
+        psample_sock = NULL;
+        return 0;
+    }
+
+    return 0;
+}
+
+static void
+dpif_offload_netlink_destroy(void)
+{
+    if (!psample_sock) {
+        return;
+    }
+
+    nl_sock_destroy(psample_sock);
+    psample_sock = NULL;
+}
+
+static void
+dpif_offload_netlink_sflow_recv_wait(void)
+{
+    if (psample_sock) {
+        nl_sock_wait(psample_sock, POLLIN);
+    }
+}
+
+static int
+psample_from_ofpbuf(struct offload_psample *psample,
+                    const struct ofpbuf *buf)
+{
+    static const struct nl_policy ovs_psample_policy[] = {
+        [PSAMPLE_ATTR_IIFINDEX] = { .type = NL_A_U16 },
+        [PSAMPLE_ATTR_SAMPLE_GROUP] = { .type = NL_A_U32 },
+        [PSAMPLE_ATTR_GROUP_SEQ] = { .type = NL_A_U32 },
+        [PSAMPLE_ATTR_DATA] = { .type = NL_A_UNSPEC },
+    };
+    struct nlattr *a[ARRAY_SIZE(ovs_psample_policy)];
+    struct genlmsghdr *genl;
+    struct nlmsghdr *nlmsg;
+    struct ofpbuf b;
+
+    b = ofpbuf_const_initializer(buf->data, buf->size);
+    nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg);
+    genl = ofpbuf_try_pull(&b, sizeof *genl);
+    if (!nlmsg || !genl || nlmsg->nlmsg_type != psample_family
+        || !nl_policy_parse(&b, 0, ovs_psample_policy, a,
+                            ARRAY_SIZE(ovs_psample_policy))) {
+        return EINVAL;
+    }
+
+    psample->iifindex = nl_attr_get_u16(a[PSAMPLE_ATTR_IIFINDEX]);
+    psample->dp_group_id = nl_attr_get_u32(a[PSAMPLE_ATTR_SAMPLE_GROUP]);
+    psample->packet = a[PSAMPLE_ATTR_DATA];
+
+    return 0;
+}
+
+static int
+psample_parse_packet(struct offload_psample *psample,
+                     struct dpif_offload_sflow *sflow)
+{
+    dp_packet_use_stub(&sflow->packet,
+                       CONST_CAST(struct nlattr *,
+                                  nl_attr_get(psample->packet)) - 1,
+                       nl_attr_get_size(psample->packet) +
+                       sizeof(struct nlattr));
+    dp_packet_set_data(&sflow->packet,
+                       (char *) dp_packet_data(&sflow->packet) +
+                       sizeof(struct nlattr));
+    dp_packet_set_size(&sflow->packet, nl_attr_get_size(psample->packet));
+
+    sflow->attr = dpif_offload_sflow_attr_find(psample->dp_group_id);
+    if (!sflow->attr) {
+        return ENOENT;
+    }
+    sflow->iifindex = psample->iifindex;
+
+    return 0;
+}
+
+static int
+dpif_offload_netlink_sflow_recv(struct dpif_offload_sflow *sflow)
+{
+    if (!psample_sock) {
+        return ENOENT;
+    }
+
+    for (;;) {
+        static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
+        struct offload_psample psample;
+        uint64_t buf_stub[4096 / 8];
+        struct ofpbuf buf;
+        int error;
+
+        ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
+        error = nl_sock_recv(psample_sock, &buf, NULL, false);
+
+        if (!error) {
+            error = psample_from_ofpbuf(&psample, &buf);
+            if (!error) {
+                    ofpbuf_uninit(&buf);
+                    error = psample_parse_packet(&psample, sflow);
+                    return error;
+            }
+        } else if (error != EAGAIN) {
+            VLOG_WARN_RL(&rl, "Error reading or parsing netlink (%s).",
+                         ovs_strerror(error));
+            nl_sock_drain(psample_sock);
+            error = ENOBUFS;
+        }
+
+        ofpbuf_uninit(&buf);
+        if (error) {
+            return error;
+        }
+    }
+}
+
+const struct dpif_offload_api dpif_offload_netlink = {
+    .init = dpif_offload_netlink_init,
+    .destroy = dpif_offload_netlink_destroy,
+    .sflow_recv_wait = dpif_offload_netlink_sflow_recv_wait,
+    .sflow_recv = dpif_offload_netlink_sflow_recv,
+};
diff --git a/lib/dpif-offload-provider.h b/lib/dpif-offload-provider.h
index c82d04e74..0a0b71618 100644
--- a/lib/dpif-offload-provider.h
+++ b/lib/dpif-offload-provider.h
@@ -17,12 +17,12 @@
 #ifndef DPIF_OFFLOAD_PROVIDER_H
 #define DPIF_OFFLOAD_PROVIDER_H
 
+#include "dp-packet.h"
 #include "netlink-protocol.h"
 #include "openvswitch/packets.h"
 #include "openvswitch/types.h"
 
 struct dpif;
-struct dpif_offload_sflow;
 
 /* When offloading sample action, userspace creates a unique ID to map
  * sFlow action and tunnel info and passes this ID to datapath instead
@@ -37,6 +37,13 @@ struct dpif_sflow_attr {
     ovs_u128 ufid;                  /* Flow ufid. */
 };
 
+/* Parse the specific dpif message to sFlow. So OVS can process it. */
+struct dpif_offload_sflow {
+    struct dp_packet packet;            /* Packet data. */
+    uint32_t iifindex;                  /* Input ifindex. */
+    const struct dpif_sflow_attr *attr; /* SFlow attribute. */
+};
+
 /* Datapath interface offload structure, to be defined by each implementation
  * of a datapath interface.
  */
@@ -61,4 +68,8 @@ void dpif_offload_sflow_recv_wait(const struct dpif *dpif);
 int dpif_offload_sflow_recv(const struct dpif *dpif,
                             struct dpif_offload_sflow *sflow);
 
+#ifdef __linux__
+extern const struct dpif_offload_api dpif_offload_netlink;
+#endif
+
 #endif /* DPIF_OFFLOAD_PROVIDER_H */
-- 
2.30.2



More information about the dev mailing list